Changelog
0.4
Chat improvements, smarter memory & context, refreshed design, and reliability fixes.
Chat
- Message queueing — type and send while the agent is still working; messages wait in a visible queue (removable chips above the composer) and are sent automatically, one by one, when the current reply finishes.
- Graceful Stop — stopping a response no longer restarts the engine. The partial reply is kept in the chat, full conversation context is preserved, and other sites' running tasks are unaffected. Esc also stops.
- Retry & Edit — hover any message for Copy; the last reply gets a Retry button, your last message gets Edit (pulls the text back into the composer).
Memory & context
- Chats survive restarts — transcripts are now saved to disk per site.
- Long-term site memory — the agent remembers durable facts (site quirks, preferences, past decisions) across sessions and new chats; stored as human-editable markdown per site.
- Smarter long conversations — replayed history is capped, so the model no longer degrades in very long chats.
Design
- Refreshed, warmer look: new color palette, serif greeting with time of day, redesigned composer with separate Stop/Send controls, subtler message bubbles, and an animated thinking indicator.
Reliability
- Interrupts are turn-scoped and race-safe (stopping one site can no longer affect another site's in-flight task).
- Partial output is preserved when a provider error occurs mid-reply.
0.3.1
- Minor bug fixes and stability improvements.
0.3
- Doubled skills (11 → 21)
- macOS native notifications
- OpenAI Codex OAuth provider
- Typing indicator
- Improved skill editor with Markdown template
0.2
- Multi-site chat management
- Skill editor UI
- Model selection settings
- Sparkle auto-update framework
0.1
- Initial public release
- 11 built-in WordPress skills
- Freemius license activation
- Sparkle updates